Data analysis using PCA method¶

Objective: To identify which handwritten digits are most distinctive from the rest using the PCA method.¶

Let's say you're working on an image recognition service for a postal company. The service wants to automatically recognize handwritten digits. (This method is already used by modern postal services and is even more accurate than humans.)
The postal company wants to know which handwritten digits are harder to recognize (relative to each other) so that it can focus on collecting more samples of that data.
You have a dataset of handwritten digits, and you need to extract their principal components using PCA and identify which digits are easy to distinguish from the rest.

Libraries you may need to use:¶

In the lower cell, load the data in the digits.csv file as a dataframe, remove the extra columns from it (number_lable column) and display it (sample output is below).¶

Select one of the images in the file (which is available as a vector in the file) and display it using the Python libraries above (your output should look something like the images below).¶

In the first step, you need to scale the data to get the correct results (so that larger values do not bias the results, you can find more information on this in this link https://shorturl.at/0V1vR ) Scale the data in the bottom cell and display it (as an array).¶

In this section, you need to find the two principal components using PCA, print the pixel coordinates accordingly (as an array), draw a scatterplot of the numbers (similar to the image below), and based on the results obtained, express your analysis in a Markdown of which numbers are more difficult to distinguish from each other.¶

Enter your analysis of the information above here.¶

In the final step, similar to the previous step, find the three principal components of PCA and find a 3D scatterplot of the numbers (similar to the image below) and express your analysis (if the graph is interactive, it has a score!)¶